imatrix : use GGUF by default

Still uses the old format when the output filename ends with .dat
but this can be overridden with --output-format

Author: compilade

common/arg.cpp

@@ -2627,6 +2627,15 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.n_out_freq = value;
         }
     ).set_examples({LLAMA_EXAMPLE_IMATRIX}));
+    add_opt(common_arg(
+        {"--output-format"}, "{gguf,dat}",
+        string_format("output format for imatrix file (default: gguf except when output filename ends with .dat)"),
+        [](common_params & params, const std::string & value) {
+            /**/ if (value == "gguf") { params.imat_out_type = COMMON_IMATRIX_FORMAT_GGUF; }
+            else if (value == "dat")  { params.imat_out_type = COMMON_IMATRIX_FORMAT_DAT;  }
+            else { throw std::invalid_argument("invalid output format"); }
+        }
+    ).set_examples({LLAMA_EXAMPLE_IMATRIX}));
     add_opt(common_arg(
         {"--save-frequency"}, "N",
         string_format("save an imatrix copy every N iterations (default: %d)", params.n_save_freq),

common/common.h

@@ -233,6 +233,12 @@ enum common_reasoning_format {
     COMMON_REASONING_FORMAT_DEEPSEEK,        // Extract thinking tag contents and return as `message.reasoning_content`, including in streaming deltas.
 };
 
+enum common_imatrix_format_type {
+    COMMON_IMATRIX_FORMAT_AUTO,
+    COMMON_IMATRIX_FORMAT_GGUF,
+    COMMON_IMATRIX_FORMAT_DAT,  // legacy
+};
+
 struct common_params {
     int32_t n_predict             =    -1; // new tokens to predict
     int32_t n_ctx                 =  4096; // context size
@@ -431,6 +437,7 @@ struct common_params {
     int32_t n_out_freq  = 10; // output the imatrix every n_out_freq iterations
     int32_t n_save_freq =  0; // save the imatrix every n_save_freq iterations
     int32_t i_chunk     =  0; // start processing from this chunk
+    common_imatrix_format_type imat_out_type = COMMON_IMATRIX_FORMAT_AUTO; // format of the output imatrix
 
     bool process_output  = false; // collect data for the output tensor
     bool compute_ppl     = true;  // whether to compute perplexity

tools/imatrix/README.md

@@ -7,7 +7,7 @@ More information is available in <https://github.com/ggml-org/llama.cpp/pull/486
 
 ```
 ./llama-imatrix \
-    -m model.gguf -f some-text.txt [-o imatrix.gguf] [--no-ppl] \
+    -m model.gguf -f some-text.txt [-o imatrix.gguf] [--output-format {gguf,dat}] [--no-ppl] \
     [--process-output] [--chunk 123] [--save-frequency 0] [--output-frequency 10] \
     [--in-file imatrix-prev-0.gguf --in-file imatrix-prev-1.gguf ...] [--parse-special] \
     [--show-statistics] [...]
@@ -20,6 +20,7 @@ The parameters in square brackets are optional and have the following meaning:
 * `-lv | --verbosity` specifies the verbosity level. If set to `0`, no output other than the perplexity of the processed chunks will be generated. If set to `1`, each time the results are saved a message is written to `stderr`. If `>=2`, a message is output each time data is collected for any tensor. Default verbosity level is `1`.
 * `-o | --output-file` specifies the name of the file where the computed data will be stored. If missing `imatrix.gguf` is used.
 * `-ofreq | --output-frequency` specifies how often the so far computed result is saved to disk. Default is 10 (i.e., every 10 chunks)
+* `--output-format` specifies the output format of the generated imatrix file. Either "gguf", or "dat" (the legacy format). Defaults to "gguf" unless the output filename ends with `.dat`.
 * `--save-frequency` specifies how often to save a copy of the imatrix in a separate file. Default is 0 (i.e., never)
 * `--process-output` specifies if data will be collected for the `output.weight` tensor. Typically, it is better not to utilize the importance matrix when quantizing `output.weight`, so this is set to `false` by default.
 * `--in-file` one or more existing imatrix files to load and combine. Useful for merging files from multiple runs/datasets.

tools/imatrix/imatrix.cpp

@@ -26,7 +26,7 @@
 static void print_usage(int, char ** argv) {
     LOG("\nexample usage:\n");
     LOG("\n    %s \\\n"
-            "       -m model.gguf -f some-text.txt [-o imatrix.gguf] [--no-ppl] \\\n"
+            "       -m model.gguf -f some-text.txt [-o imatrix.gguf] [--output-format {gguf,dat}] [--no-ppl] \\\n"
             "       [--process-output] [--chunk 123] [--save-frequency 0] [--output-frequency 10] \\\n"
             "       [--in-file imatrix-prev-0.gguf --in-file imatrix-prev-1.gguf ...] [--parse-special] \\\n"
             "       [--show-statistics] [...]\n" , argv[0]);
@@ -492,13 +492,15 @@ void IMatrixCollector::save_imatrix_legacy(int32_t ncall) const {
 
 void IMatrixCollector::save_imatrix(int32_t n_chunk) const {
     auto fname = m_params.out_file;
+    auto imat_type = m_params.imat_out_type;
 
-    // TODO: use the new format in more cases
-    if (!string_ends_with(fname, ".gguf")) {
-        LOG_WRN("\n%s: saving to legacy imatrix format because output suffix is not .gguf\n", __func__);
+    if ((imat_type == COMMON_IMATRIX_FORMAT_AUTO && string_ends_with(fname, ".dat")) ||
+        (imat_type == COMMON_IMATRIX_FORMAT_DAT)) {
+        LOG_WRN("\n%s: saving to legacy imatrix format\n", __func__);
         this->save_imatrix_legacy(n_chunk);
         return;
     }
+    // else, default to GGUF imatrix
 
     if (n_chunk > 0) {
         fname += ".at_";